The present invention relates to a training method for an echo canceller for use in a voice conference system.
Voice conference systems, which have become known in the art, are effective for holding a conference or meeting of participants or attendants living in remote locations because the participants need not be physically present in the conference or meeting. Such systems are used to connect a plurality of auditoriums by a wire, a radio, and/or a satellite communication network.
A voice conference system uses a loudspeaker as a receiver and a microphone used as a transmitter. Since the loudspeaker is unavoidably acoustically coupled to the microphone, it is inevitable that a part of the audible sound reproduced by the loudspeaker is picked up by the microphone, resulting in a reverberation signal. The reverberation signal is sent back to the remote party, and in an unfavorable case, the reverberation signal gives rise to howling.
In order to eliminate echoes in the room without intercepting the voice signal of neither the transmitting or the receiving party, it is effective to provide an echo canceller within the voice conference system. The echo canceller consists of an adaptive filter for generating a simulated echo signal according to an input voice signal, a subtractor for subtracting the simulated echo signal from the input voice signal and generating a residual echo signal, and a filter coefficient updating circuit for updating the coefficients to reduce the residual echo signal to zero. In such a conventional echo canceller for use in a voice conference system, the response characteristic of the adaptive filter in its initial state is different that of the echo path, so that the echo canceller should be trained before the conference to bring the former close to the latter. A white noise signal is usually used for the training of the canceller so that the echo cancellation can be achieved substantially uniformly within the transmission band.
Conventionally, an echo canceller for a voice conference system is trained every time the line is to be connected to the remote party, by inputting a training signal to the loudspeaker and the echo canceller before the start of the conference. This is disadvantageous in that the echo canceller should be trained every time a remote party is newly connected and reproduced noise comes out of the loudspeaker to annoy the conferees in the conference room.